Typically, a computer system contains a processor, a bus, and other peripheral devices. The processor is responsible for executing instructions using data in the computer system. The bus is used by the processor and the peripheral devices for transferring information between one another. The information on the bus usually includes data, address and control signals. The peripheral devices comprise storage devices, input/output (I/O) devices, etc. Generally, all operations being performed in the computer system occur at the same frequency.
Many of today's computer systems include power management capabilities. Power management is used to reduce the dynamic and static power consumption of a system to increase the battery life of a mobile personal computer (PC) or to reduce the energy costs associated with a desktop PC. Dynamic power is consumed by all components during state switching of internal electronic circuits, while static power is consumed due to the leakage currents of electronic devices.
The existing power-management techniques in a typical notebook and desktop PC use specific hardware mechanisms to provide maximum power savings. These hardware mechanisms use processor specific interrupts (e.g., System Management Interrupt (SMI)) and other system activity monitoring hardware (e.g., idle timers and hardware trapping mechanisms) to provide a reasonable amount of power conservation.
Alternatively, some software mechanisms exist, which are used today to detect CPU idle conditions in order to put the system in an optimum power conservation mode (e.g., Windows APM driver, DOS POWER.EXE, etc.). Although the available software techniques provide for about seventy to eighty percent of power conservation, they do not power manage a system beyond CPU idleness. That is, they do not detect idleness of I/O devices nor turn off idle I/O devices during system operation or slow the CPU clock rate, etc.
In existing power management architecture's, there are several dynamic and static power conservation states. One of these states is referred to as a fully on or full power on state, in which all the components of a typical system are powered. In this state, all the clocks in the system will be running at full speed. This state offers no power savings. Another state is referred to as a local stand-by, or partially powered-on, state, in which certain temporarily idle local devices in the system, such as a floppy device, graphics devices (e.g., LCD, CRT), hard disk device, etc. are powered down. The power to these turned off devices is restored when an internal or external system event requires the services of these resources. The system maintains idle timers for each of these power-manageable devices. The idle timers enter an expired time-out state when they detect idleness of these devices after a pre-defined period of inactivity, and notifies the power management software. This state offers the minimal amount of power savings in a system. Another state is referred to as a global stand-by state, in which most of the system devices are powered down with the exception of the CPU and the system DRAM memory. The clock to the CPU is stopped with the DRAM memory operating in an extended power conservation mode, sometimes referred to as stand-by mode with self refresh. At this point in time, the CPU and DRAM are ready to be activated when a system event occurs. An example of such a system event is a keyboard/mouse click or other system interrupts (e.g., IRQ0-IRQ15, NMI, SMI, etc.). The last power management state is referred to as hibernation, where the system is put in the power-off state. When a system detects an idle condition, after a predetermined period of time in the global stand-by mode, it can initiate a transfer to the hibernation state. In such a state, complete system state is saved to the hard disk. When the system is turned back on, the hibernation state restores the system back to exactly the same state as it was before.
Dynamic clock throttling is the state where the dynamic power consumed by a CPU is reduced by slowing its clock rate. A slow clock to the CPU is emulated by periodic assertion and de-assertion of a stpclk (stop clock) signal. This slow clock emulation leads to less power consumption by the overall system. This mode is activated during normal operation of a system and it is overlapped with a fully-on state to offer additional power savings during the fully-on state.
As shown above, the prior art system of power management requires the detection of local and global system events. This is typically handled by special hardware or power-aware applications and device drivers. The detection of system idleness for global stand-by is accomplished by monitoring their interrupt activity in software or hardware using existing methods. If none of the system interrupts are activated in a predetermined of time, a global stand-by event is generated by the chosen hardware or software mechanism. Local activity of individual I/O devices is detected mostly by dedicated power management hardware. The hardware snoops on I/O device resources (e.g., I/O addresses and IRQx, etc.). The mapping of the resources is static and deterministic and known at system boot-up time. These I/O resource mappings do not change over the lifetime of the current system boot. In certain power management implementations, the snoop I/O addresses are programmable in the I/O hardware, while they are fixed in other systems. The deterministic nature of the mappings of these I/O resources of the local devices (as per PC-AT/DOS standards) makes it easy to design standard hardware which is consistent across all PC DOS platforms.
These described methodologies have several inherent problems. For instance, each of the I/O devices needs an idle timer to monitor the activity. This imposes a restriction on a number of hardware timers that can be designed into the system. Also, most implementations hardcode the I/O trapping address of the I/O devices to save "gates". This makes a system more sensitive to remapping of the I/O resources. Furthermore, the existing mechanisms assume that all I/O devices use standard I/O resource mappings over the lifetime of the system, i.e., static I/O and IRQx mapping. This, in fact, places a severe restriction on the usage of the system resources and demands perfect hardware compatibility across all platforms.
Power management software in the traditional system is completely decoupled from the operating system and application. This makes the system prone to the operating system and power management software performing activities, with neither of them being aware of the activities being performed by the other. This may lead to system crashes where a power management interrupt, such as SMI, takes control away from the operating system while it is executing in the middle of a critical section of code.
The current generation of device drivers in operating systems virtualize I/O ports. When the I/O ports are virtualized, it becomes difficult, and in some cases impossible, for the power management software and hardware to detect any possible remappings. This leads the power management hardware to monitor and trap on invalid I/O device addresses, thereby generating improper events in the system.
In a plug-and-play environment, it is assumed that the I/O device resource mappings (I/O and IRQx) are no longer deterministic or visible to the power management software at system boot-up time. Current and future generations of operating systems will be based on plug-and-play architectures, where the I/O resource mapping can and will change dynamically during the lifetime of the current system boot. When these dynamic remappings of I/O devices do occur, there is currently no easy way to communicate to the power management hardware and software.
Also, the existing software power-management techniques assume that the applications of the associated device drivers in the system are APM and PMC aware. This makes it difficult to manage a system with applications and drivers which are not power aware.